AI Incident Response — Part 1: Declare, Contain, Preserve & Communicate — 45 Templates, Checklists & Worksheets for Declaring, Severity, the First Hour, Containment, Evidence & Disclosure

The dashboard was green in every tile while the assistant told people the wrong thing all night. 45 production templates, checklists & worksheets distilled from Chapters 1–7 of AI Incident Response: the harm-based incident definition and four triage questions, the four-class severity matrix with the exposure override, the first hour in four fifteen-minute blocks, the containment ladder and vendor pin, the thirty-minute evidence capture routine, the notification cascade, and disclosure outside the building — customers, the public, and the regulator pattern — ending in the book's 61-line Part 1 gate. PDF + Markdown source.

The dashboard was green in every tile. The harm ran all night anyway.

An AI incident rarely announces itself. The model is up, latency is normal, the error rate is a flat and reassuring zero — while the system fluently gives people wrong answers they act on, or shows one person's information to another. The first hour then goes to a group chat debate about whether it "counts" as an incident, nobody declares anything so nobody owns the timeline, and a helpful engineer rebuilds the very state investigators will need. AI Incident Response — Part 1 is the machinery that replaces that hour: 45 copy-paste templates, checklists, and worksheets distilled from Chapters 1–7 of the dedicated incident-response book — decide what deserves the alarm, set severity as a decision with clocks attached, run the first hour in four fifteen-minute blocks, stop the bleeding with containment levers that work on behaviour rather than infrastructure, preserve the scene before the cleanup destroys it, and communicate inside and outside the building on one timeline, in plain words, on the awareness clock.

Every item carries the book's tested positions built in: an incident defined by harm, not by error messages — four branches, any one enough — sorted from its neighbours (defect, outage, near miss, complaint) by four triage questions asked in a fixed order, with "unknown" counted toward declaration; a four-class severity matrix read across as a package of clock, cadence, and notifications, scored on five dimensions where the highest sets the class and never gets averaged away, with one rule outranking the scoring — a suspected cross-person exposure starts high and steps down only on evidenced scoping; six first-hour roles with exactly one name on each, a change freeze with snapshot-before-change, and a commander checklist carried whole; a containment ladder that shapes first, rolls back one named component at a time, pins the vendor in writing so no silent update rewrites your incident mid-response, purges source, index, cache, session, and downstream in order, and verifies by harm to people over a stated window — never by error rate; an evidence discipline that captures model identity, running configuration, and retrieval state in perishability order, sealed with hashes under a chain of custody and a plain-language hold, in a rehearsed thirty minutes; and a communication discipline — cascade in rings of need, inside before outside, updates whose next-update time is always stated and always kept, an executive brief that names decisions instead of topics, and disclosure outside built on six principles, affected-population rings enumerated from evidence, and separate instruments for direct notice and public statement. Where the book gives an example — its severity clocks, its fifteen-minute blocks, its capture routine — the template labels it as the book's example, illustrative for a mid-size operator, for you to refill with your own committed values. Bracketed fields are yours to fill; nothing here invents a figure for your organisation. The book's worked incident — Lakeshore Health Network, its assistant Clara, its drafter Scribe, its vendor Copperline AI — is a fictional composite, labeled fictional in the book itself. This pack is incident-response machinery, exactly as the book frames it: it is not legal advice, it gives no regulatory-filing guidance, and it guarantees no outcome.

What's inside — 45 templates, checklists & worksheets in 8 sections

1. An Incident Is Not a Bug — Deciding What Deserves the Alarm (6) — Harm-Based Incident Definition Sheet · The Neighbours Sorting Table · Why AI Incidents Hide — Four Concealment Checks & the Behavioural Signal Layer · Four Triage Questions Funnel Worksheet · Raise vs Declare — Rights, Authority & the Quiet-Downgrade Ban · The Price of the Wrong Call — Asymmetry Sheet & First-Hour Diagnosis

2. Severity Is a Decision — Classes, Triggers & the Courage to Declare (6) — Four-Class Severity Matrix · Five-Dimension Severity Scoring Worksheet · The Exposure Override — Suspected Data Exposure Starts High · What Each Class Buys — Clocks, Cadence, Notifications & Records · Declare Early, Downgrade on Facts — the Operating Rule & Its Countermeasures · Worked Classification — Lakeshore: SEV-1 at About 04:20, SEV-2 the Next Morning

3. The First Hour — Roles, Rhythm & the First Sixty Minutes (6) — Six Roles Assignment Sheet · Four Fifteen-Minute Blocks Run Sheet · One Channel, One Timeline, One Voice — the First-Hour Triad · Change Freeze & Snapshot-Before-Change Rule Sheet · Incident Commander Checklist · Handover & Second-Hour Staffing Sheet

4. Stop the Bleeding — Containment When the Failure Is a Behaviour (6) — Harm-Path Framing Sheet · The Containment Ladder & Lever Comparison Table · Rollback Card — Four Components to Last-Known-Good · Disable & Isolate Sweep Sheet · Vendor Pin Protocol — No Silent Updates During an Incident · Purge Order & Harm Verification — with the Worked Containment of Two Systems

5. Preserve the Scene — Evidence Before It Evaporates (7) — Why AI Evidence Dies Fast — the Five Forces & Your Real Deadlines · Capture List I — Model Identity, Running Configuration & Retrieval State · Capture List II — Samples, Clocks & the Response's Own Change Log · Evidence Capture Table · Evidence Manifest & Chain of Custody Sheet · Legal Hold in Plain Language & Protecting the Evidence Itself · The Thirty-Minute Capture Routine

6. Inside the Building — Stakeholders, Executives & the Rhythm of Updates (6) — Notification Cascade Table — Inside Before Outside, in Rings of Need · Cadence by Severity — the Chapter 6 Rhythm & the Three Protective Rules · Internal Update Template — Status, Impact, Actions, Next Update · Executive Brief — a Decision Instrument, One Page · Front-Line Words & Script Versioning Sheet · Rumor Control & the Single Timeline — Be the Fastest Accurate Source

7. Outside the Building — Customers, the Public & the Regulator (6) — Six Disclosure Principles Checklist · Affected-Population Rings & the Boundary Decision Record · Direct Notice vs Public Statement — Decision Table & Sequencing Rule · Customer Notice Anatomy & Skeleton · Regulator Notification as an Operational Pattern — Awareness Clock, Partial-by-Design Initial Notice & Skeleton · The Vendor's Place in Your Disclosure & the Two-Notices Worked Example

8. Daylight Decisions & the Part 1 Gate (2) — Daylight Readiness Build Sheet — every decision the chapters demand in advance, consolidated · Part 1 First-Response Gate — Chapters 1–7 Sign-Off — the book's seven chapter checklists consolidated into one evidence-backed gate (61 lines)

Who this is for

Incident commanders and on-call leads who hold the phone when an AI system misbehaves, platform and reliability leads who own containment levers and evidence, communications and privacy staff who own the cascade, the disclosure wording, and the awareness clock, and the executives who get the early-morning call and must decide in minutes. The source book assumes you operate at least one AI system that touches customers, staff, money, or records, and that your organisation already has some incident habit for ordinary software. If your team's first hour would go to debating whether it is "really an incident," start at Template 1.

What this pack is not

Not Part 2. Eradication and staged recovery, the post-incident review, the corrective-action register, tabletops and game days, and on-call design and response metrics are the book's Chapters 8–12 — and Appendices A (Master Checklist), B (Quick Reference), and C (Template Library) travel with them. They live in the Part 2 companion pack; nothing here restates them, and this pack is complete and useful without them. Where Chapters 1–7 point forward to that material, this pack carries the pointer only.

Not legal advice, and not regulatory-filing guidance. The source book's own position, carried in full: notification duties, regulatory clocks, and disclosure obligations vary by jurisdiction, sector, and contract, and must be confirmed with qualified counsel for your situation before being relied upon. This pack builds the response machinery — declaration, containment, evidence, communication — and carries regulator notification only as an operational readiness pattern (awareness timestamp, facts feed, skeleton held ready). It does not tell you what you must file, with whom, or by when, and it guarantees no disclosure or response outcome.

Not the observability pack. The companion AI Observability pack detects and monitors AI systems in production — telemetry, tracing, evaluation, drift, SLOs — and holds that book's incident runbook item. This pack is a different book and a different discipline: it begins where detection ends, at the decision to declare, and runs the response. The two packs share no template.

Not the cross-book security slice. The companion AI Security Templates & Checklists pack holds the MCP-and-enterprise-agents security artifacts, including its own agent incident response plan for agent systems. This pack is the dedicated response book's machinery for AI incidents generally — classification, severity, first hour, containment, evidence, disclosure — and ships no security control, audit-trail standard, or vendor risk assessment.

Not the governance machinery. The companion AI Governance in Practice — Part 1 pack builds the standing portfolio machinery — register, tiering, approval lanes, policy, cards. This pack governs nothing standing; it responds to the night the machinery is not enough.

No platform or tooling. The timeline format, evidence manifest, and notice skeletons are simplified, structurally honest templates. No incident platform, paging tool, or evidence store ships here.

Companion products

The AI Observability pack is the detection-and-monitoring companion — the machinery that notices. The AI Security Templates & Checklists pack is the cross-book security companion. AI Governance in Practice — Part 1 is the standing-governance companion. AI Incident Response — Part 2: Recovery, Post-Incident Review & Readiness is this pack's sibling, built from Chapters 8–12 of the same book plus Appendices A, B, and C: a companion for when the first response is built and you are ready for the recovery, review, and readiness half — never a requirement for this pack to work. This pack stands alone.

Note: the $99 Architecture Templates Bundle currently contains Products #1–#3 only. This pack launches standalone; no bundle inclusion is promised for this product.

Built by the author of AI Incident Response: The Runbook for Model Failures, Leaks, and Attacks, from the incident-response discipline behind production enterprise AI systems.

About the author

Naveen Kumar Josyula is the author of 35 AI books on Amazon KDP and the CEO and founder of AISoltrics Technologies Inc. He is the author of the source book, AI Incident Response: The Runbook for Model Failures, Leaks, and Attacks, from which every template in this pack is distilled.

Support promise

If anything in this file is wrong or unclear, email support@aisoltrics.com before you judge it — we fix verified issues within one business day and send you the corrected file free. (You can also reach the founder's office at aadhya@aisoltrics.com.) Updates to this product are free to past buyers via your Payhip library. Sold by AISoltrics Technologies Inc.